Measurement and Analysis of TCP-Friendly Rate Control for Vertical Handovers

An intersystem or vertical handover is a key enabling mechanism for next generations of mobile communication systems. A vertical handover can cause an abrupt change of up to two orders of magnitude in link bandwidth and latency. It is hard for end-to-end congestion control to adapt promptly to such changes. This is especially a concern for slowly responsive congestion control algorithms such as TFRC designed to provide a smooth transmission rate for real-time applications and therefore less responsive to changes in network conditions than TCP. In the first part of the paper, we measure performance of TFRC and TCP flows during a vertical handover between GPRS, WLAN, and LAN in our testbed based on Mobile IP. As expected, TFRC is less aggressive than TCP over fast links. However, after a handover to a slow link TFRC can starve a competing TCP flow. In the second part of the paper, we use a model of an ideal handover between UMTS, GPRS and WLAN. An ideal handover does not cause error losses, delays, packet duplication or reordering. These negative effects can be partly or completely eliminated by new handover algorithms; an abrupt change in latency and link bandwidth after a vertical handover cannot be avoided. As expected, TFRC can overshoot or underutilize the new link significantly after a handover. Tuning TFRC parameters including self-clocking, history discounting and higher feedback frequency from the receiver has a positive but insufficient effect on TFRC responsiveness and aggressiveness. We evaluate two new mechanisms to improve transport performance for vertical handovers: overbuffering and an explicit handover notification. Overbuffering prevents packet losses during handovers between links with a different bandwidth-delay product and is more useful for TCP than to TFRC. The explicit handover notification allows to instantly adapt the TFRC rate on a new link, which otherwise can take up to a minute.